Statistical block-based skin detection

In this paper, we propose a novel approach to model the human skin color. The underlying approach involves affecting a block's average value at each pixel location, using its surrounding points. The generated skin data are found to follow a generalized Gaussian distribution (GGD) and a mixture of GGDs in the H and S color component, respectively. Next, the model parameters are estimated using the maximum-likelihood (ML) criterion applied to a set of training skin samples. Each pixel is then classified as skin or the opposite if its joint likelihood ratio is above some threshold. The preliminary experimental results show that our model avoids excessive false detection while still retaining a high degree of correct detection.